Bing and Copilot: Getting Into Microsoft's AI Answers

Microsoft Copilot, the AI summaries on Bing results pages, and Azure's Grounding with Bing Search all draw on one store: the Bing index. Microsoft documents Web Search and Grounding with Bing Search as generally available and Bing-index-backed in its Azure agent docs (updated July 2026). If Bing does not hold your page, those answers cannot cite it.

It helps to split the path into four stages, because you control them very differently:

1. Parametric. What the underlying model knows with no search at all. It is frozen at a training cutoff. Nobody changes it this quarter.
2. Access. Whether bingbot can fetch your page and Bing can store it. This is yours to fix.
3. Retrieval. Whether your page comes back for the searches Copilot runs to answer a question. Copilot writes its own search phrases rather than sending the user's prompt word for word.
4. Generation. Which retrieved pages the model actually uses and cites. The model decides.

Microsoft framed the shift in May 2026: "search indexing was built to help humans decide what to read. Grounding indexing is being built to help AI systems decide what to say." The practical reading is that classic Bing ranking and AI grounding overlap but are not the same system. For a deeper walk through the pipeline, see how Copilot grounds answers in Bing. If you also care about Google's AI surfaces, the Bing index and the Google index feed different AI answers, and health in one tells you nothing about the other.

Bing Webmaster Tools launched an AI Performance report in public preview on 10 February 2026 and expanded it on 16 June 2026. It covers Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations. It reports:

• Total Citations and Average Cited Pages
• Page-level citation activity, so you can see which URLs get used
• Grounding queries: Microsoft's definition is "the key phrases the AI used when retrieving content that was referenced in AI-generated answers"
• Intents and Topics for those queries
• Citation Share and a Compare view

Grounding queries are the unusual part. Every prompt-tracking tool tests questions its users chose. Grounding queries are the searches the AI actually ran. As of October 2026, no other engine publishes anything equivalent. Expect them to surprise you: Copilot often cites pages for queries you never targeted, because it rewrites and splits the user's question.

Three caveats matter when you read the report:

• Microsoft says the grounding query data "represents a sample of overall citation activity." That note sits on the grounding query metric, not the whole report.
• Microsoft calls Citation Share "an observational metric, not a ranking system or a competitive scoreboard." It can fall while your absolute citations rise, because whoever already wins the query dominates the denominator.
• As of the evidence base's August 2026 check, the report has no API. You export it by hand.

If you have not verified your site yet, read whether Bing Webmaster Tools is worth setting up for AI search. For a routine that turns exports into decisions, see tracking Copilot grounding queries.

A common pitch says ChatGPT uses Bing, so getting into Bing gets you into ChatGPT. The evidence does not support selling that as a measured path.

Here is what OpenAI actually says. Its help centre states that ChatGPT search "sometimes partners with other search providers" and rewrites your question into one or more targeted queries it sends to them. "Sometimes" is the operative word. The provider list in circulation traces back to 2024-era documentation, and nobody has published an end-to-end measurement of how Bing inclusion affects ChatGPT citations.

The one public test that looked points the other way. In OtterlyAI's 2026 experiment, two new sites each published about 1,000 AI-generated posts. Google deindexed both. Bing kept both, and combined Bing impressions reached 119,935 in July. Of the 3,934 AI citations Otterly tracked for one site, 2,718 (69%) came from Microsoft Copilot. ChatGPT citations were effectively zero, even while ChatGPT's crawlers visited and the sites ranked in Bing.

That is one pair of sites on a low-competition topic, run by a tracking vendor. It does not prove Bing never helps ChatGPT. It does show that Bing presence alone did not produce ChatGPT citations in the only published case. Our write-up of the Otterly experiment's method and limits covers what else it can and cannot support.

The honest line: Bing is a documented upstream of Copilot and Azure grounding, and a plausible but unquantified input to ChatGPT. ChatGPT search has its own gate. OpenAI's crawler docs say sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers," so check that bot separately.

The levers with evidence behind them are access levers, and they work by removing blocks. None of them guarantees a citation.

On topical fit: a July 2026 survey of 45 GEO studies found topical relevance and context position were the most reproducible factors. It also concluded that no technique shows a stable, cross-platform causal effect on discoverability. Relevance means covering the subject your buyers ask about. It does not mean keyword variants or rewriting for an AI style, which the evidence does not support.

One on-page move has a small positive signal: state specific facts (a price, a number, a one-line definition) in plain HTML on a page Bing can fetch. That gives the engine a sentence it can quote. It is not a promise that it will.

Work from the gate inward. Each step is cheap, and a failure early makes every later step pointless.

1. Check robots.txt for bingbot. Make sure no blanket rule shuts it out. A minimal permissive entry looks like this:

User-agent: bingbot
Allow: /

2. Check the firewall. A CDN or bot-management rule can return 403 to a crawler that robots.txt allows, and nothing in your analytics will say so. Request a page as bingbot and confirm a 200 with real content, not a challenge page. The free AI Crawler Checker does this per crawler. More detail is in bingbot, Copilot and your firewall.
3. Verify the site in Bing Webmaster Tools. Confirm your key pages are indexed. This is the only way to see Bing index status.
4. Search for suppression tags. Bing added data-nosnippet support on 15 October 2025. Content marked with it "is still indexed normally, but it will be excluded from snippets and AI summaries." Templates often apply it to more than a cookie banner. Also check noindex, nosnippet and max-snippet:0. See whether data-nosnippet keeps content out of Copilot.
5. Export the AI Performance report. List your grounding queries and cited pages. Note which buyer topics never appear.
6. Put load-bearing facts in the HTML. Pricing, product names and plan limits should be readable without running JavaScript and consistent across pages, so Copilot has one correct sentence to quote.
7. Check what Copilot says. Citations show where it looked, not whether it got your facts right. Ask it the questions buyers ask and compare the answers with your own pages. The free checker on this site runs a snapshot of that for up to three buyer questions on other engines too.

• Reading Citation Share as a leaderboard. Microsoft says it is not one. Track absolute citations and cited pages alongside it.
• Treating IndexNow as an AI lever. IndexNow is a free way to tell Bing a URL changed. The specification never mentions AI, and a 200 response confirms only that the ping was received. Microsoft asserts it helps AI reference current pages, but no study backs that.
• Assuming Google health covers Bing. In Otterly's test the two indexes diverged completely. Check each webmaster tool on its own.
• Selling Bing as the ChatGPT strategy. ChatGPT search depends on OAI-SearchBot access and OpenAI's own discretion. Check that gate separately.
• Adding markup to get cited. No markup, phrasing or page structure has credible evidence of getting a page cited. Existing schema does no harm. It just is not a lever for AI answers.
• Mass-publishing thin pages. Copilot citing a deindexed capybara site does not make up for losing Google, which removed both test sites without a manual action notice.
• Trusting one prompt tracker's count. Microsoft's report credited one Otterly site with 42,817 citations against 3,934 in Otterly's tracker. The windows may differ, but any prompt set is a sample.

Can Copilot cite a page that Google has deindexed? Yes, if Bing still holds it. In OtterlyAI's 2026 experiment, Google removed both test sites, Bing kept them, and Microsoft Copilot produced 69% of one site's tracked citations. Google's own AI surfaces almost never cited them after removal. The lesson is that each engine can only cite what its own index holds, so check Google Search Console and Bing Webmaster Tools separately.

Does submitting URLs through IndexNow get me into Copilot answers faster? Nothing measured says so. IndexNow notifies Bing and six other endpoints that a URL changed. Google does not consume it. The protocol documentation never mentions AI, and a successful response only confirms receipt. Microsoft has said it helps AI systems reference current pages, but published no study. Use it as free Bing housekeeping, not as a citation tactic.

Does data-nosnippet remove my page from Bing's index? No. Bing's October 2025 announcement says content marked with data-nosnippet is still indexed normally and stays available for ranking, but it is excluded from snippets and AI summaries. That makes it a display control. It is useful for hiding a specific block, and harmful when a template applies it to your pricing or product copy by accident.

Is there an API for Bing's grounding query data? Not as of the evidence base's August 2026 check. The Bing Webmaster API reference shows no citation or grounding endpoints, so the AI Performance report has to be read or exported by hand. Microsoft has signalled that API access may arrive, but plan around manual exports for now and keep the raw files so you can compare months.

Will allowing bingbot guarantee Copilot citations? No. Allowing bingbot buys eligibility to be crawled. After that, Bing decides what to index, Copilot decides which searches to run, and the model decides which retrieved pages to cite. You control access, reachability and suppression tags. You do not control retrieval or citation, and nobody can promise you a place in an answer.